Adding relative spatial coordinate maps or using early FiLM-style modulation with convolutions, rather than attention or relation modules, drives performance on simple spatial language in VQA.
The visual QA devil in the details: The impact of early fusion and batch norm on CLEVR
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
What is needed for simple spatial language capabilities in VQA?
Adding relative spatial coordinate maps or using early FiLM-style modulation with convolutions, rather than attention or relation modules, drives performance on simple spatial language in VQA.